Time series data analysis device, time series data analysis method, and computer program

Abstract

According to one embodiment, a time series data analysis device includes a feature vector calculator and an updater. The feature vector calculator calculates feature amounts of a plurality of feature waveforms based on distances between a partial time series and the feature waveforms, the partial time series being data belonging to each of a plurality of intervals which are set in a plurality of pieces of time series data. The updater updates the feature waveforms based on the feature amounts.

1 . A time series data analysis device comprising:
 a feature vector calculator configured to calculate feature amounts of a plurality of feature waveforms based on distances between a partial time series and the feature waveforms, the partial time series being data belonging to each of a plurality of intervals which are set in a plurality of pieces of time series data; and   an updater configured to update the feature waveforms based on the feature amounts.   
     
     
         2 . The time series data analysis device according to  claim 1 , wherein a set of the intervals entirely covers the time series data. 
     
     
         3 . The time series data analysis device according to  claim 1 , further comprising a feature waveform selector configured to set the intervals by repeatedly specifying a pair of an interval and any one of the feature waveforms in a certain range from an interval that is set immediately before, the pair achieving a minimum distance between the any one of the feature waveforms and a partial time series in the interval of the pair. 
     
     
         4 . The time series data analysis device according to  claim 1 , wherein the feature vector calculator specifies a plurality of the partial time series having a minimum distance with any one of the feature wave forms and calculates a feature amount of the any one of the feature waveforms based on a maximum distance among distances between the any one of the feature wave forms and the plurality of the partial time series. 
     
     
         5 . The time series data analysis device according to  claim 1 , wherein the updater calculates gradients of the feature waveforms and updates the feature waveforms based on the gradients. 
     
     
         6 . The time series data analysis device according to  claim 1 , wherein the updater updates a model parameter of a one-class identifier based on the feature amounts by a gradient method. 
     
     
         7 . The time series data analysis device according to  claim 6 , wherein the one-class identifier is an evaluation formula which includes input variables representing the feature amounts and the model parameter. 
     
     
         8 . The time series data analysis device according to  claim 6 , wherein the one-class identifier is a linear or non-linear one-class SVM. 
     
     
         9 . The time series data analysis device according to  claim 6 , further comprising an anomaly detector, wherein
 the feature vector calculator calculates feature amounts of a plurality of second feature waveforms as the updated feature waveforms based on distances between a partial time series in each of a plurality of intervals set in time series data as a test target and the second feature waveforms, and   the anomaly detector determines whether the time series data as a test target has anomaly based on the model parameter and the feature amounts of the second feature waveforms.   
     
     
         10 . The time series data analysis device according to  claim 9 , wherein
 the feature vector calculator specifies a plurality of the partial time series having a minimum distance with any one of the second feature waveforms and calculates a feature amount of the any one of second feature waveforms based on a maximum distance among distances between the any one of the second feature waveforms and the plurality of the partial time series, and   the time series data analysis device further includes an anomaly specifier configured to specify, when the anomaly is detected in the time series data as a test target, a distance between the partial time series in each interval in the time series data as a test target and the second feature waveform having the minimum distance from the partial time series among the second feature waveforms, compare the specified distance with the maximum distance of the second feature waveform, and determine the partial time series for which the distance is larger than the maximum distance to be an anomaly waveform.   
     
     
         11 . The time series data analysis device according to  claim 1 , wherein
 a plurality of ranges are set in the time series data,   a plurality of feature waveforms are specified in the ranges, and   the feature vector calculator calculates the feature amounts based on a minimum distance between the partial time series in each interval and one of a plurality of feature waveforms specified in the range to which each interval belongs.   
     
     
         12 . The time series data analysis device according to  claim 11 , wherein
 the time series data is obtained by connecting pieces of time series data of variables in a temporal direction, and   the feature waveforms are specified per each of ranges corresponding to the piece of time series data of the variables.   
     
     
         13 . A time series data analysis method comprising:
 calculating feature amounts of a plurality of feature waveforms based on distances between a partial time series and the feature waveforms, the partial time series being data belonging to each of a plurality of intervals which are set in a plurality of pieces of time series data; and   updating the feature waveforms based on the feature amounts.   
     
     
         14 . A non-transitory computer readable medium having a computer program stored therein which causes a computer to perform processes comprising:
 calculating feature amounts of a plurality of feature waveforms based on distances between a partial time series and the feature waveforms, the partial time series being data belonging to each of a plurality of intervals which are set in a plurality of pieces of time series data; and   updating the feature waveforms based on the feature amounts.